To solve the cost burden caused by delivery tardiness for small-and-medium-sized discrete manufacturing enterprises, an assembly-oriented three-layer planning and scheduling framework is proposed based on the multi-agent technology. Taking the advantage of the autonomy and coordination of multi-agent, and combining with heuristic rules, this framework makes the complex large-scale production planning decomposed into more easily solvable sub-problems with optimization algorithms. The resulted plans of every layer are updated iteratively based on the down layer optimization till the near-optimal solution is achieved. Moreover, with the planning execution information, dynamic scheduling can also be addressed within this framework. The simulation with the historical data of gravure printing machine orders demonstrates the capabilities of the scheme for the tardiness penalty problem through the effective control of the idle-time distribution.